Daily Stock Prediction Using Neuro-genetic Hybrids

We propose a neuro-genetic daily stock prediction model. Traditional indicators of stock prediction are utilized to produce useful input features of neural networks. The genetic algorithm optimizes the neural networks under a 2D encoding and crossover. To reduce the time in processing mass data, a parallel genetic algorithm was used on a Linux cluster system. It showed notable improvement on the average over the buy-and-hold strategy. We also observed that some companies were more predictable than others.